HUMAN PERCEPTION AND COMPUTER EXTRACTION OF BEAT STRENGTH

Musicalsignalsexhibit periodictemporalstructurethatcreates the sensationof rhythm. In orderto model,analyze,andretrieve musicalsignalsit is importantto automaticallyextract rhythmic information. To somewhat simplify the problem,automaticalgorithms typically only extract informationaboutthe main beatof thesignalwhich canbelooselydefinedastheregularperiodicsequenceof pulsescorresponding to wherea humanwould tap his foot while listeningto the music. In thesealgorithms,thebeatis characterizedby its frequency (tempo),phase(accentlocations) andaconfidencemeasureabout its detection. Themain focusof this paperis the conceptof Beat Strength, which will be looselydefinedasthe rhythmic characteristicsthat allow to discriminatebetweentwo piecesof musichaving thesame tempo. Using this definition, we can say that a pieceof Hard Rockhasa strongersenseof rhythmthana ClassicalMusic at the sametempo. Characteristicsrelatedto Beat Strength have been implicitely usedin automaticbeatdetectionalgorithmsandshown to be as importantas tempoinformation for musicclassification andretrieval. In this work, a userstudyexploring the perception of Beat Strength wasconductedandtheresultswereusedto calibrateandexplore automaticBeat Strength measuresbasedon the calculationof Beat Histograms.